AI Agents Advance in Security and Enterprise Adoption, Amidst Emerging Risks

Key Intelligence

• ■OpenAI introduced "Codex Security" in research preview, an AI agent tool for context-aware vulnerability detection, validation, and patch generation across codebases.
• ■Anthropic's Claude Opus 4.6 AI model successfully identified 22 vulnerabilities in Firefox, highlighting AI's potential in cybersecurity research.
• ■PLDT partnered with UiPath to launch "ERICA," an agentic AI service specifically designed for enterprise risk management, indicating increasing business adoption of agentic AI.
• ■Discussions emphasize that successfully deploying AI agents to production requires more than just better models, and that the concept of AI "hallucinations" is often misunderstood.
• ■Microsoft warned that threat actors are actively operationalizing AI for malicious tradecraft, underscoring the critical need for advanced AI security measures.